notesum.ai
Published at December 5Learning Speed-Adaptive Walking Agent Using Imitation Learning with Physics-Informed Simulation
cs.RO
cs.LG
Released Date: December 5, 2024
Authors: Yi-Hung Chiu1, Ung Hee Lee2, Changseob Song1, Manaen Hu1, Inseung Kang1
Aff.: 1Carnegie Mellon University; 2University of Michigan

| Reward Ratio | Target vs. Measured Speed | Ground Truth vs. Measured Joint Angle | |||
| Discriminator | Speed | RMSE (m/s) | RMSE (°) | ||
| 1 | 0 | 0.510.02 | -0.850.14 | 5.420.16 | 0.800.01 |
| 0.9 | 0.1 | 0.190.01 | 0.730.02 | 3.860.07 | 0.900.01 |
| 0.8 | 0.2 | 0.130.01 | 0.880.01 | 4.970.06 | 0.850.01 |
| 0.7 | 0.3 | 0.080.02 | 0.950.02 | 5.380.15 | 0.810.01 |
| 0.6 | 0.4 | 0.070.01 | 0.960.01 | - | - |
| 0.5 | 0.5 | 0.060.01 | 0.980.01 | 5.240.08 | 0.830.01 |